Model Context Protocol (MCP) server for WebRTC stats

What the rtcStats MCP server does

Point an MCP-compatible agent at rtcStats and it can pull analyzed session data through natural language: list the sessions that failed, open one, and read the observations, experience score, and AI summary, without leaving the editor. On the Free plan you do the same thing by uploading the dump in the browser; the MCP-in-IDE path is available on the Developer plan and up.

Tools

• •get_quota - Returns the application token account: plan, allowed credits, and remaining credits (monthly plus purchased pool).
• •list_sessions - Lists analyzed sessions with rich filters (browser, OS, experience score band, observation types and tags, user/conference/session IDs).
• •get_session - Returns the full analysis for one session: summary, observations, session features, experience score, and the plain-English AI summary.

Connect your agent

The server speaks Streamable HTTP at https://api.rtcstats.com/v1.0/mcp. Send your application token as Authorization: Bearer on the request. Create a token under Settings > Applications (Developer plan).

Choose Read-only as the scope. All three tools above are reads, so a read-only token runs every one of them, and the same token is refused with 403 insufficient_scope on the endpoints that spend a credit or delete a session (upload, analyze, enrich and DELETE /v1.0/sessions/{id}). It is worth doing: this token sits in a config file on disk, and config files get committed and pasted into issues. Create a full-access token instead if you want the same credential to upload or analyze dumps.

Claude Desktop / Claude Code

{
  "mcpServers": {
    "rtcstats": {
      "type": "http",
      "url": "https://api.rtcstats.com/v1.0/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_APPLICATION_TOKEN"
      }
    }
  }
}

Related

• •How to debug a webrtc-internals dump with AI
• •All integrations (REST API, MCP, embeddable viewer)
• •Pricing and plans (MCP is on the Developer plan)